Title: High-dimensional causal inference that capitalizes on experimental design and computing, illustrated with epigenomics
Authors: Marie-Abele Bind - Massachusetts General Hospital (United States) [presenting]
Abstract: In a randomized experiment, no matter how unorthodox the design or the basic statistical analysis, a valid $p$-value is available. This fact had to be obvious to RA Fisher before 1925, and was recognized for its potential practical utility by Brillinger, Jones, and Tukey (in the context of cloud seeding experiments decades ago in 1978). The practical utility of the idea should be even more evident today because of the widespread availability of high-speed computing and non-traditional statistical methods for data analysis (such as Lasso-based methods), which also rely on high-speed computing to complete in realistic time. Despite the simplicity of the argument, the use of this approach seems to be relatively recondite in current statistical practice. This article attempts to rectify this lacunae through the use of a simple example of a small randomized experiment with a high-dimensional epigenetic outcome.